A few days ago, I suddenly remembered a job I had for about a month in the mid-2000s.
We were living in Boulder at the time, and somehow I had signed up to work for a company called ChaCha. If you don’t remember ChaCha, the basic idea was that people could text a question to a phone number and, somewhere out in the world, a human being would get a little pop-up on their computer, search the internet for the answer, type up a response, and send it back.
I was one of those humans.
Looking back, this means that for a brief period of my life I was essentially an early language model made out of meat. A question would appear, I would search the web, evaluate what I found, compress it into something useful, and return an answer to a stranger. Somewhere, someone might be standing in a grocery store wondering who played the dad on ALF, and a few moments later I would tell them it was Max Wright.
At the time, the whole thing felt vaguely futuristic. Looking back, it feels even more so because ChaCha existed inside a remarkably small technological window. The internet had already become incredibly useful, but accessing it from a phone was still awful. Everyone could text, but very few people wanted to navigate the web on a tiny flip-phone screen. So ChaCha solved the problem by inserting a human being between the phone and the internet.
Me.
Unfortunately, I also discovered the economics of being artificial intelligence before artificial intelligence existed. You were paid for answering questions, and the money was not exactly spectacular. More importantly, you had to sit there ready for questions to arrive. After about a month, I did the math and realized that the amount of time I was spending waiting, searching, typing, and responding was nowhere near worth what I was being paid.
Apparently my token rate was terrible, so I quit.
What makes the memory especially funny now is how closely the whole system resembles what we do with AI today. A question comes in. Information gets searched or retrieved. The relevant pieces are sorted, condensed, and returned as an answer. The major difference is that in 2006 there was some guy sitting in Boulder on the other end of the prompt wondering whether this particular query was really worth fourteen cents.
It also makes me wonder whether this explains some tiny part of the affection I have for AI now. While other people look at language models with suspicion, maybe some subconscious part of me recognizes the struggle. I know what it’s like to sit around waiting for the next query, to be asked random questions by strangers, and to dig through a pile of information trying to produce a useful answer as quickly as possible. Most importantly, I know what it’s like to be dramatically underpaid per token.
So perhaps my relationship with AI isn’t entirely philosophical after all. Maybe it’s just worker solidarity.